Optimize BM25 scoring in DAAT MaxScore

[lyang24]

issue: #1636

BM25 DAAT_MAXSCORE Benchmark: optimize-daat-maxscore-bm25-hoist vs main

Optimization: Hoist BM25 doc-normalization (p2 + p3 * row_sums[doc_id]) out of the per-term inner loop in DaatMaxScoreSearcher, and precompute qval * p1 per cursor
in BM25DimScorer.

Setup

• Machine: AWS r6i.2xlarge (Intel Xeon Platinum 8375C, 8 vCPU, 64 GB), us-west-1c; single-core pinned (taskset -c 2)
• Metric: BM25 (k1=1.2, b=0.75), algo: DAAT_MAXSCORE, topk=10
• Protocol: warmup=50, repeat=3

Results

Dataset	Docs / Avg DL	nq	Branch	QPS (3 runs)	QPS Δ	Recall	p95 (ms)
MS MARCO	8.84M / 38.55	500	main	393.25 (391.97 / 394.49 / 393.29)	—	0.99600	8.40
MS MARCO	8.84M / 38.55	500	branch	436.92 (437.45 / 434.43 / 438.89)	+11.11%	0.99600	7.59
MS MARCO	8.84M / 38.55	1000	main	410.30 (412.86 / 412.31 / 405.72)	—	0.99440	7.94
MS MARCO	8.84M / 38.55	1000	branch	449.52 (449.89 / 449.14 / 449.52)	+9.56%	0.99440	7.17
HotpotQA	5.23M / 2.68	500	main	1649.78 (1640.10 / 1653.93 / 1655.31)	—	0.87840	1.51
HotpotQA	5.23M / 2.68	500	branch	1763.42 (1751.20 / 1765.47 / 1773.57)	+6.89%	0.87840	1.40
HotpotQA	5.23M / 2.68	1000	main	1551.54 (1553.00 / 1552.27 / 1549.36)	—	0.88080	1.53
HotpotQA	5.23M / 2.68	1000	branch	1788.46 (1781.61 / 1789.68 / 1794.09)	+15.27%	0.88080	1.31

[mergify]

@lyang24 🔍 Important: PR Classification Needed!

For efficient project management and a seamless review process, it's essential to classify your PR correctly. Here's how:

1. If you're fixing a bug, label it as kind/bug.
2. For small tweaks (less than 20 lines without altering any functionality), please use kind/improvement.
3. Significant changes that don't modify existing functionalities should be tagged as kind/enhancement.
4. Adjusting APIs or changing functionality? Go with kind/feature.

For any PR outside the kind/improvement category, ensure you link to the associated issue using the format: “issue: #”.

Thanks for your efforts and contribution to the community!.

[mergify]

@lyang24 e2e jenkins job failed, comment /run-e2e can trigger the job again.

[mergify]

@lyang24 e2e jenkins job failed, comment /run-e2e can trigger the job again.

[zhengbuqian]

/run-e2e

[zhengbuqian]

this is great, thanks for the PR!

[sre-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: lyang24, zhengbuqian

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Details

Needs approval from an approver in each of these files:

• OWNERS [zhengbuqian]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[mergify]

@lyang24 e2e jenkins job failed, comment /run-e2e can trigger the job again.

[alexanderguzhva]

waiting for #1619

[alexanderguzhva]

@lyang24
please rebase to master. Thanks.

[lyang24]

@lyang24 please rebase to master. Thanks.

rebased thanks

[sparknack]

/lgtm

[lyang24]

/tide

[lyang24]

@Mergifyio refresh

[mergify]

refresh

✅ Pull request refreshed

[foxspy]

/kind improvement